The virtual slice setup
Introduction
Neural simulation has traditionally been practiced more like American rather than like international football: discrete simulations followed by regrouping and reorganization to prepare for the next attempt. This method necessarily distances the practice of simulation from in vivo neurophysiology, where experiments are performed on an active dynamical system which is never truly statistically stationary. It is more similar to performing experiments on a quiescent brain slice that requires repeated shocks to produce transient activity but again dissimilar to slice experiments on an active, firing network—an “epileptic” slice.
An alternative to the traditional simulation method has been called reactive animation (RA) by Efroni et al., 2005, Efroni et al., 2007. The “reactive” refers to reactive systems, a term originating in engineering and now being introduced in biology. In engineering, reactive systems can be distinguished from transformational systems, which are designed to terminate in a distinct output. Reactive systems, by contrast, operate in real-time (e.g., cruise controls and autopilots) and produce outputs that are state dependent. A particular output is only correct when it is produced at the correct time: the reactive system is in a continuous dynamical interplay with its environment. Seen in these terms, all biological systems are reactive systems. A biological system is continuously evolving, reacting to inputs that may also alter the system itself (plasticity). As with aircraft or process engineering, a reactive, real-time, ongoing biological system may be best served by use of reactive simulation.
The “animation” of reactive animation is obligatory rather than cosmetic: it provides the means for interaction with the running simulation, providing continuous or statistical evaluation of state variables and allowing control of system parameters. Like a video game, the quality of the simulation experience depends largely on the adherence to both the pragmatics and the dynamics of the system. As we will show, the experience of immediate interaction with the simulation can lead one to make improvements to this realism. However, neurophysiological simulation still suffers from a severe lack of detail compared to engineering systems or to simulation of other organ systems. In particular, there is a lack of detailed wiring information for brain areas, contrasting markedly with the relatively sophisticated knowledge of the single neuron.
Compared to experiment, simulation offers advantages of detailed observability and control. One has the ability to see all voltages and concentrations and to manipulate any neurotransmitter or ion channel at will. Indeed, one of the difficult problems in designing an RA simulator is adapting the graphical environment to the user, showing the user necessary information for a particular experiment without overwhelming him with extraneous data or multiple control panels.
Although the formalized notions of RA are relatively new to biology, the idea of interactive simulation in neurophysiology dates back at least to P. Rowat’s “Preparation” simulator. This lobster stomatogastric ganglion simulator was developed in the late 1980s, only about 5 years after the development of stand-alone graphical workstations made sophisticated graphics readily available (Rowat and Selverston, 1993). More recently, M. Hereld and collaborators have been advancing the idea of interactive simulations running on large parallel supercomputers in continuous communication with a front-end graphical workstation (Hereld et al., 2007). The virtual slice (VS) setup developed here has the advantage of being fairly large (expandable to about neurons on a standard workstation) without requiring a supercomputer. Here we illustrate a 2700 cell simulation which runs at approximately 2 model minutes/hour on a laptop. This simulation rate makes it easy to run ion channel and synaptic blockade experiments over periods of several seconds of simulated time.
Section snippets
Materials and methods
The techniques and simulations described here are implemented in the NEURON simulator web site, 2007, Carnevale and Hines, 2006 using a rule-based artificial cell mechanism Lytton and Hines, 2004, Lytton and Stewart, 2005, Lytton and Stewart, 2006. This neuron model is a fast event-driven unit that was designed with several of the attributes of biological neurons, including adaptation, bursting, depolarization blockade, Mg++-sensitive NMDA conductance, anode-break depolarization, and others.
The
Using the virtual slice setup
The default display for the VS consists of only two windows, though with multiple recording sites and parameter panels the display can quickly grow to encompass many tens of panels. Because the program is profligate in its use of windows, they are grouped so as to allow sets of them to be closed or reopened in concert. The two default windows are population activity display and the control panel (Fig. 1). The activity display (left) shows the total number of spikes occurring over a customizable
Advantages of continuous simulation
An advantage of the VS is that usage of this tool tends to suggest further improvements to allow the VS to conform still more closely to experimental physiology practice. This is best illustrated by the Wash in button near the top of the weight parameters panel (Fig. 5A) which was not originally included. The discrepancy with experimental practice was noticeable when clicking the Change weights button which effected instantaneous alterations in synaptic strengths. Such instantaneous effects are
Discussion
We have designed, developed and demonstrated an interactive neuronal network simulation with several attributes reminiscent of experimentation with an electrophysiological slice preparation: the ability to place electrodes to record either intracellularly or by selected population; a facility for placing stimulation electrodes at particular locations and stimulating with variable strength; the ability to wash-in and wash-out neuroactive ligands with the potential for multiple effects on
Acknowledgment
This research is supported by NIH (NS045612 and NS11613).
References (18)
- et al.
Discrete event simulation in the neuron environment
Neurocomputing
(2004) - et al.
Rule-based firing for network simulations
Neurocomputing
(2006) - et al.
Simulation of large networks: technique and progress
- et al.
The NEURON book
(2006) - et al.
Reactive animation: realistic modeling of complex dynamic systems
Computer
(2005) - et al.
Emergent dynamics of thymocyte development and lineage determination
Plos Comput Biol
(2007) - et al.
Framework for interactive million-neuron simulation
J Clin Neurophysiol
(2007) - et al.
Hybrid neural networks–combining abstract and realistic neural units
IEEE Eng Med Biol Soc Proc
(2004) - et al.
A rule-based firing model for neural networks
Int J Bioelectromagn
(2005)
Cited by (12)
Towards a real-time interface between a biomimetic model of sensorimotor cortex and a robotic arm
2014, Pattern Recognition LettersCitation Excerpt :In the current model we trained synaptic weights between spiking units using global reward and punisher signals. Individual neurons were modeled as rule-based dynamical units with several key features found in real neurons, including adaptation, bursting, depolarization blockade, and voltage-sensitive NMDA conductance (Lytton and Stewart, 2005; Lytton and Stewart, 2006; Lytton and Omurtag, 2007; Lytton et al., 2008; Lytton et al., 2008; Neymotin et al., 2011b; Kerr et al., 2012). The model consisted of 384 excitatory and 128 inhibitory cells, each with three types of synaptic inputs commonly found in cortex (AMPA, NMDA and GABAA).
Restoring behavior via inverse neurocontroller in a lesioned cortical spiking model driving a virtual arm
2016, Frontiers in NeuroscienceCortical spiking network interfaced with virtual musculoskeletal arm and robotic arm
2015, Frontiers in NeuroroboticsEpilepsy: Neocortical simulation for epilepsy surgery guidance: Localization and intervention
2014, Computational Surgery and Dual Training: Computing, Robotics and ImagingMotor cortex microcircuit simulation based on brain activity mapping
2014, Neural Computation